1. Field of the Invention
This invention relates to data compression, specifically to the compression and decompression of video images.
2. Description of Prior Art
In the last few years, there have been tremendous advances in the speed of computer processors and in the availability of bandwidth of worldwide computer networks such as the Internet. These advances have led to a point where businesses and households now commonly have both the computing power and network connectivity necessary to have point-to-point digital communications of audio, rich graphical images, and video. However the transmission of video signals with the full resolution and quality of television is still out of reach. In order to achieve an acceptable level of video quality, the video signal must be compressed significantly without losing either spatial or temporal quality.
A number of different approaches have been taken but each has resulted in less than acceptable results. These approaches and their disadvantages are disclosed by Mark Nelson in a book entitled The Data Compression Book, Second Edition, published by M&T Book in 1996. Mark Morrision also discusses the state of the art in a book entitled The Magic of Image Processing, published by Sams Publishing in 1993.
Video Signals
Standard video signals are analog in nature. In the United States, television signals contain 525 scan lines of which 480 lines are visible on most televisions. The video signal represents a continuous stream of still images, also known as frames, that are fully scanned, transmitted and displayed at a rate of 30 frames per second. This frame rate is considered full motion. A television screen has a 4:3 aspect ratio.
When an analog video signal is digitized, each of the 480 lines is sampled 640 times and each sample is represented by a number. Each sample point is called a picture element, or pixel. A two dimensional array is created that is 640 pixels wide and 480 pixels high. This 640×480 pixel array is a still graphical image that is considered to be full frame. The human eye can perceive 16.7 thousand colors. A pixel value comprised of 24 bits can represent each perceivable color. A graphical image made up of 24 bit pixels is considered to be full color. A single second-long full frame, full color video requires over 220 millions bits of data.
The transmission of 640×480 pixels×24 bits per pixel times 30 frames requires the transmission of 221,184,000 millions bits per second. A T1 Internet connection can transfer up to 1.54 millions bits per second. A high speed (56 Kb) modem can transfer data at a maximum rate of 56 thousand bits per second. The transfer of full motion, full frame, full color digital video over a T1 Internet connection, or 56 Kb modem, will require an effective data compression of over 144:1, or 3949:1, respectively.
Basic Run-Length Encoding
An early technique for data compression is run-length encoding where a repeated series of items are replaced with one sample item and a count for the number of times the sample repeats. Prior art shows run-length encoding of both individual bits and bytes. These simple approaches by themselves have failed to achieve the necessary compression ratios.
Variable Length Encoding
In the late 1940s, Claude Shannon at Bell Labs and R. M. Fano at MIT pioneered the field of data compression. Their work resulted in a technique of using variable length codes where codes with low probabilities have more bits, and codes with higher probabilities have fewer bits. This approach requires multiple passes through the data to determine code probability and then to encode the data. This approach also has failed to achieve the necessary compression ratios.
D. A. Huffman disclosed a more efficient approach of variable length encoding known as Huffman coding in a paper entitled “A Method for Construction of Minimum Redundancy Codes,” published in 1952. This approach also has failed to achieve the necessary compression ratios.
Arithmetic, Finite Context, and Adaptive Coding
In the 1980s, arithmetic, finite coding, and adaptive coding have provided a slight improvement over the earlier methods. These approaches require extensive computer processing and have failed to achieve the necessary compression ratios.
Dictionary-Based Compression
Dictionary-based compression uses a completely different method to compress data. Variable length strings of symbols are encoded as single tokens. The tokens form an index to a dictionary. In 1977, Abraham Lempel and Jacob Ziv published a paper entitled, “A Universal Algorithm for Sequential Data Compression” in IEEE Transactions on Information Theory, which disclosed a compression technique commonly known as LZ77. The same authors published a 1978 sequel entitled, “Compression of Individual Sequences via Variable-Rate Coding,” which disclosed a compression technique commonly known as LZ78 (see U.S. Pat. No. 4,464,650). Terry Welch published an article entitled, “A Technique for High-Performance Data Compression,” in the June 1984 issue of IEEE Computer, which disclosed an algorithm commonly known as LZW, which is the basis for the GIF algorithm (see U.S. Pat. Nos. 4,558,302, 4,814,746, and 4,876,541). In 1989, Stack Electronics implemented a LZ77 based method called QIC-122 (see U.S. Pat. Nos. 5,532,694, 5,506,580, and 5,463,390). The output of a QIC-122 encoder consists of a stream of data, which, in turn consists of tokens and symbols freely intermixed. Each token or symbol is prefixed by a single bit flag that indicates whether the following is a dictionary reference or a plain symbol. The definitions for these two sequences are:                (a) plaintext: <1><eight-bit-symbol>        (b) dictionary reference: <0><window-offset><phrase-length>Windows offsets are encoded as seven bits or eleven bits. These lossless (method where no data is lost) compression methods can achieve up to 10:1 compression ratios on graphic images typical of a video image. While these dictionary-based algorithms are popular, these approaches require extensive computer processing and have failed to achieve the necessary compression ratios.JPEG and MPEG        
Graphical images have an advantage over conventional computer data files: they can be slightly modified during the compression/decompression cycle without affecting the perceived quality on the part of the viewer. By allowing some loss of data, compression ratios of 25:1 have been achieved without major degradation of the perceived image. The Joint Photographic Experts Group (JPEG) has developed a standard for graphical image compression. The JPEG lossy (method where some data is lost) compression algorithm first divides the color image into three color planes and divides each plane into 8 by 8 blocks, and then the algorithm operates in three successive stages:                (a) A mathematical transformation known as Discrete Cosine Transform (DCT) takes a set of points from the spatial domain and transforms them into an identical representation in the frequency domain.        (b) A lossy quantization is performed using a quantization matrix to reduce the precision of the coefficients.        (c) The zero values are encoded in a zig-zag sequence (see Nelson, pp. 341-342).        
JPEG can be scaled to achieve a higher compression ratio by allowing more loss in the quantization stage of the compression. However this loss results in certain blocks of the image being compressed such that areas of the image have a blocky appearance and the edges of the 8 by 8 blocks become apparent because they no longer match the colors of their adjacent blocks. Another disadvantage of JPEG is smearing. The true edges in an image get blurred due to the lossy compression method.
The Moving Pictures Expert Group (MPEG) uses a combination of JPEG based techniques combined with forward and reverse temporal differencing. MPEG compares adjacent frames and for those blocks that are identical to those in a previous or subsequent frame and only a description of the previous or subsequent identical block is encoded. MPEG suffers from the same blocking and smearing problems as JPEG.
These approaches require extensive computer processing and have failed to achieve the necessary compression ratios without unacceptable loss of image quality and artificially induced distortion.
QuickTime: CinePak, Sorensen, H.263
Apple Computer, Inc. released a component architecture for digital video compression and decompression, named QuickTime. Any number of methods can be encoded into a QuickTime compressor/decompressor (codec). Some popular codec are CinePak, Sorensen, and H.263. CinePak and Sorensen both require extensive computer processing to prepare a digital video sequence for playback in real time; neither can be used for live compression. H.263 compresses in real time but does so by sacrificing image quality resulting in severe blocking and smearing.
Fractal and Wavelet Compression
Extremely high compression ratios are achievable with fractal and wavelet compression algorithms. These approaches require extensive computer processing and generally cannot be completed in real time.